Methods and Apparatus to Label Radiology Images

1. An apparatus, comprising: means for obtaining user input identifying a vertebra on a spinal image; means for generating first annotations on the spinal image based on: 1) the user input; 2) connected regions on the spinal image; and 3) a number of viewable vertebrae on the spinal image, the connected regions on the spinal image being determined based on contextual-information features that account for the user input, the contextual-information features including information about the viewable vertebrae or information about neighboring structures, the number of viewable vertebrae being determined from the contextual-information features; means for identifying an error in the first annotations; means for generating second annotations on the spinal image based on: 1) the error; 2) second connected regions on the spinal image; and 3) the number of viewable vertebrae on the spinal image, the second connected regions on the spinal image being determined based on second contextual-information features that account for the error; means for displaying the spinal image including the second annotations; means for validating the second annotations in association with the spinal image; and means for storing the second annotations in association with the spinal image.

2. The apparatus of claim 1 , wherein the contextual-information features are generated based on feature vectors for points in an image domain.

3. The apparatus of claim 2 , further including: means for generating a model using the feature vectors and associated with the user input; means for identifying the contextual-information features that are consistent with the model; means for identifying centroids of the connected regions on the spinal image of the contextual-information features that are consistent with the model; and means for selecting first points that form a shape of a spine on the spinal image from the centroids.

4. The apparatus of claim 3 , wherein the means for identifying the error in the first annotations include means for identifying an error in the shape of the spine.

5. The apparatus of claim 4 , wherein the error in the shape of the spine includes a vertebra on the spinal image not corresponding to one of the first annotations.

6. The apparatus of claim 4 , wherein the means for generating the second annotations includes means for selecting second points that correct for the error in the shape of the spine relative to the first points, the second points including second user input identifying the error in the first annotations.

7. The apparatus of claim 6 , wherein the second user input corresponds to a coordinate vector of a point in the image domain.

8. The apparatus of claim 2 , wherein the means for generating the second annotations includes means for maximizing a shape-based criterion associated with the points based on second user input identifying the error in the first annotations.

9. The apparatus of claim 8 , wherein the shape-based criterion includes an angle constraint associated with vertebrae alignment on the spinal image and a distance constraint associated with vertebrae spacing.

10. The apparatus of claim 8 , wherein the means for maximizing the shape-based criterion associates distances between adjacent vertebrae on the spine that are approximately the same in accordance with the shape-based criterion.

11. A system, comprising: a processor to: identify a vertebra on a spinal image; generate first annotations on the spinal image based on: 1) user input; 2) connected regions on the spinal image; and 3) a number of viewable vertebrae on the spinal image, the connected regions on the spinal image being determined based on contextual-information features that account for the user input, the contextual-information features including information about the viewable vertebrae or information about neighboring structures, the number of viewable vertebrae being determined from the contextual-information features; identify an error in the first annotations; generate second annotations on the spinal image based on: 1) the error; 2) second connected regions on the spinal image; and 3) the number of viewable vertebrae on the spinal image, the second connected regions on the spinal image being determined based on second contextual-information features that account for the error; display the spinal image including the second annotations; validate the second annotations in association with the spinal image; and store the second annotations in association with the spinal image.

12. The system of claim 11 , wherein the contextual-information features are generated based on feature vectors for points in an image domain.

13. The system of claim 12 , wherein the processor is to generate a model using the feature vectors and associated with the user input, and wherein the processor is to identify the contextual-information features that are consistent with the model.

14. The system of claim 13 , wherein the processor is to identify centroids of the connected regions on the spinal image of the contextual-information features that are consistent with the model.

15. The system of claim 14 , wherein the processor is to optimize a cost function including a first constraint associated with a Bhattacharyya measure of similarity between feature distributions and a second constraint associated with removing regions associated with image noise.

16. A tangible machine-readable storage medium comprising instructions which, when executed, cause a processor to at least: obtain user input identifying a vertebra on a spinal image; generate first annotations on the spinal image based on: 1) the user input; 2) connected regions on the spinal image; and 3) a number of viewable vertebrae on the spinal image, the connected regions on the spinal image being determined based on contextual-information features that account for the user input, the contextual-information features including information about the viewable vertebrae or information about neighboring structures, the number of viewable vertebrae being determined from the contextual-information features; identify an error in the first annotations; generate second annotations on the spinal image based on: 1) the error; 2) second connected regions on the spinal image; and 3) the number of viewable vertebrae on the spinal image, the second connected regions on the spinal image being determined based on second contextual-information features that account for the error; display the spinal image including the second annotations; validate the second annotations in association with the spinal image; and store the second annotations in association with the spinal image.

17. The tangible machine-readable storage medium of claim 16 , wherein the contextual-information features are generated based on feature vectors for points in an image domain.

18. The tangible machine-readable storage medium of claim 17 , wherein the feature vectors include image statistics associated with image patches of different orientations or scales.

19. The tangible machine-readable storage medium of claim 18 , wherein the image patches encode the information about the viewable vertebrae or the information about the neighboring structures.

20. The tangible machine-readable storage medium of claim 19 , wherein the information includes at least one of size, shape, orientation, or relationships to neighboring structures.